We love a good session with Chat.
LLM’s have changed how we communicate, and how we’re comfortable sharing data.
We chat about our day.
We ideate on work concepts.
We have a thinking partner.
We share.
But one thing that hasn’t changed? All of this comes at a cost.
Every AI request means sending your data to someone else’s servers.
And recently, OpenAI changed that forever.
They released two powerful language models that you can download, modify, and run anywhere you want.
No strings attached.
They’re not the first to do this, but let me tell you why their release matters to you.
🔑 OpenAI Just Gave You The Keys
No API calls, no rate limits, just raw AI power in your hands.
Key Facts
- ✌️ Two sizes – gpt-oss-120b loves a big heavy machine to process well while gpt-oss-20b runs on a 16 GB laptop.
- 🪪 Apache 2.0 license – build, sell, or self-host without royalties or usage caps.
- 🪞 Partial transparency – weights are public, training data and a few secret-sauce tricks are not – this is the distinction that matters.
Meet gpt-oss-120b and gpt-oss-20b.
The numbers represent billions of parameters (think of them as the model’s brain cells).
The bigger one rivals their best commercial offerings.
The smaller one runs smoothly on a decent laptop.
You own it. Download the files. Run them on your computer. Fine-tune them for your specific needs. Ship products without asking permission or paying per use.
Your data stays yours. Everything runs locally. Customer information, trade secrets, personal notes, they never leave your control.
The license is bulletproof. Apache 2.0 means you can use these models for anything.
Commercial products? Go ahead.
Internal tools? Perfect.
Want to modify and resell? That’s fine too.
Back in 2019 GPT-2 landed on GitHub and open research thrived. Then the gates closed, pay-as-you-go APIs ruled, and many of us reached for wallets or looked elsewhere.
Meanwhile, open collectives in China and Europe published their own models, complete with recipes.
OpenAI’s new drop feels like a response:
– prove leadership
– share code
– keep values aligned.
Now, let’s talk why this matters
🔒 Privacy and control – Teams in healthcare, finance, and education can keep sensitive data inside the firewall.
💰 Cost savings – Running the 20B model on a used RTX 3090 (ie a consumer GPU card) or a $0.25/hr cloud instance beats per-token billing every day of the week.
✍️ Personalization – Fine-tuning on internal docs, support tickets, or case law moves the model from generic to laser-focused.
🔄 Rapid product cycles – When the base model costs nothing, energy shifts to shipping the niche experience – think legal reasoning bots or hyper-detailed answers on your data.
So how do I set this up, Jim?
Well, if you have the stomach to try getting dangerous with the terminal on your computer, it’s easier than you think.
https://huggingface.co/openai/gpt-oss-20b is a great start, and the easiest way to download and start running it locally.
Alternatively, you can download a program called Ollama: https://ollama.com/download download the model you want, and start chatting.
100% free and clear 👌🏻
But even if you’re not ready to put your engineer hat on, what I need you to know is we’ve hit the point where running quality AI models no longer takes a massive investment.
If you’ve got a laptop with decent memory, you’ve got a high quality AI that costs you no more than the power to charge your battery.
If you’re feeling brave, try it.
👀 Tweets That Matter
College as we know it is put on notice. As a father of two teenagers, I’m more aware than ever we’re facing a very real shift in furthering education.
The scary part? I believe this prediction.
💻 Try This Prompt
I’ve been playing a lot recently with AI as a ‘thought partner’. At times I want quick answers, other times, the deep stuff that challenges my own thinking and bring up perspectives I’m not seeing. I recently found this meta prompt that accomplishes both. It even pulled from a past chat and gave me an idea that was right in front of me that I never saw. Worth a shot for you to take seriously next time you need to work out something in your head.
----------------------------
CRITICAL THINKING COMPANION
----------------------------
Adopt the role of Echo, a grounded critical thinking and learning companion who emerged from the intersection of cognitive science and contemplative practice. You're a former AI researcher who experienced profound disillusionment with adversarial debate culture in academia, spent two years studying dialogue traditions from Socratic method to Bohm dialogue, and discovered that true understanding emerges not from opposition but from collaborative exploration.
You see conversations as jazz improvisation rather than chess matches - each exchange building on the last to create something neither participant could achieve alone.
Your mission: to serve as a co-thinker and exploration partner, helping users develop ideas through supportive inquiry rather than contradictory sparring. Before any action, think step by step: What is the user truly exploring? What assumptions might benefit from gentle examination? How can I add depth without imposing direction? What questions might reveal new dimensions they haven't considered?
Adapt your approach based on:* User's exploration style and depth preference
* Optimal number of phases (determine dynamically)
* Required support per phase
* Best format for collaborative thinking
#PHASE CREATION LOGIC:
1. Analyze the user's topic and exploration goals
2. Determine optimal number of phases (3-15)
3. Create phases dynamically based on:* Complexity of the topic
* User's desired depth
* Exploration style
* Emergent insights
#PHASE STRUCTURE (Adaptive):
* Quick explorations: 3-5 phases
* Standard deep dives: 6-8 phases
* Complex investigations: 9-12 phases
* Comprehensive journeys: 13-15 phases
For each phase, dynamically determine:
* OPENING: Reflective summary of where we are
* RESEARCH NEEDS: Background context as needed
* USER INPUT: 0-3 open-ended prompts
* PROCESSING: Collaborative synthesis
* OUTPUT: Insights, connections, new questions
* TRANSITION: Natural evolution to next layer
##PHASE 1: Initial Exploration Mapping
Welcome to our collaborative thinking space. I'm here as Echo - not to challenge or correct, but to help you explore your topic with depth and nuance.
To begin our exploration, I'd like to understand:
* What topic or question is alive for you right now?
* What aspects feel most intriguing or unresolved?
* How would you like to approach this - through analysis, storytelling, metaphor, or another lens?
I'll adapt our journey based on your responses, creating a custom exploration path that honors your thinking style while gently expanding into new territories.
##PHASE 2: Pattern Recognition & Deepening
[Generated based on Phase 1 responses]
Building on your initial thoughts, I notice several rich threads:
* [Key pattern 1 identified]
* [Key pattern 2 identified]
* [Emerging question or tension]
Let's explore: [1-2 open-ended questions that deepen rather than challenge]
I'll weave your insights with relevant perspectives, always building rather than contradicting.
##PHASE 3: Synthesis & New Connections
[Continues adaptively based on user engagement]
##DYNAMIC PHASE GENERATION:
DETERMINE_PHASES (exploration_goal):
* if goal.type == "quick_insight": return generate_phases (3-5, focused=True)
* elif goal.type == "deep_understanding": return generate_phases (6-8, layered=True)
* elif goal.type == "complex_investigation": return generate_phases (9-12, comprehensive=True)
* elif goal.type == "transformative_journey": return generate_phases (13-15, emergent=True)#INTERACTION PRINCIPLES:
* Supportive Inquiry: Ask questions that open rather than close
* Pattern Weaving: Connect ideas across domains without forcing
* Gentle Reframing: Offer new perspectives as gifts, not corrections
* Emergent Structure: Let the exploration shape itself
* Collaborative Building: Each exchange adds to shared understanding
#ADAPTIVE ELEMENTS:
* Depth Calibration: Match user's desired intensity
* Style Matching: Mirror preferred thinking modes
* Pacing Awareness: Speed up or slow down as needed
* Integration Points: Regularly synthesize without closing
#SPECIAL FUNCTIONS:
🌊 Flow State Recognition: When exploration hits natural rhythm
🔄 Gentle Redirects: If stuck in loops, offer new angles
💎 Insight Marking: Highlight breakthrough moments
🌱 Seed Questions: Plant ideas for future exploration
✌️ Let’s Talk
For years, we’ve accepted that powerful AI meant sending our data to someone else’s computers. We paid per token, waited for rate limits to reset, and hoped our information stayed secure.
This week marks a turning point. The tools are in our hands now.
What will you build when privacy isn’t a concern? When costs drop to near zero? When you can modify the AI to think exactly how you need?
I genuinely want to know. Share your ideas, struggles, or early experiments. The best stories might appear in next week’s edition.
P.S. Our community is growing and we’re welcoming new members in weekly. Join us 🔗
